CN104574413A - Blood vessel bifurcation extracting method and system of lung CT picture - Google Patents

Abstract

The invention belongs to the field of picture processing, and provides a blood vessel bifurcation extracting method and system of a lung CT picture. The spherical characteristic of the blood vessel bifurcation on a geometrical structure and the accumulation performance of tensor voting are fully used, spherical tensor voting is carried out on pixel points, belonging to a blood vessel, on the reinforced lung CT picture, the partial maximum value of a spherical tensor significance coefficient is solved to obtain an alternative blood vessel bifurcation, then the tail end points of the blood vessel are removed through a main component analysis algorithm, and an accurate lung blood vessel bifurcation is obtained. The method and system do not need earlier stage operation of the picture, has high robustness on noise, provides important basis for later picture registration and lung movement, is favorable for later lung CT picture analysis, and assists in clinical diagnosis and treatment.

Description

A kind of vascular bifurcation point extracting method of lung CT image and system

Technical field

The invention belongs to image processing field, particularly relate to a kind of vascular bifurcation point extracting method and system of lung CT image.

Background technology

Lung CT image is the reconstruction image utilizing the different scanning of tissue to the absorption of X ray and transmitance to obtain.The unique point extracting lung CT image has important application meaning, is prerequisite and the basis of visual processes and image procossing.The feature of lung CT image mainly contains angle point, marginal point, flex point, border and texture etc., in lung CT image, vascular bifurcation point, because of its special geometry, is the basis of task in the computer vision fields such as lung estimation, image registration, has a direct impact the precision of the automatic Calibration in graphical analysis and registration.

The vascular bifurcation point extracting method of the lung CT image that prior art proposes can be roughly divided into three classes: the first kind is the bifurcation extracting method based on vessel centerline, the general flow of these class methods is, the blood vessel structure of lung images is first extracted with blood vessel segmentation algorithm, carry out skeletal extraction to obtain single pixel blood vessel tree structure with blood vessel thinning algorithm again, and the node extracted from vascular tree is as vascular bifurcation point.This algorithm depends on the precision of blood vessel segmentation and the accuracy of refinement in earlier stage, easily produces many flase drop bifurcations, or there is undetected situation to weak blood vessel; Equations of The Second Kind is the bifurcation extracting method based on Hessian matrix, first these class methods want computed image matrix of second derivatives to be also Hessian matrix, then solution matrix eigenwert, builds a wave filter with Feature Combination, to obtain peak response at bifurcation place.These class methods, to noise-sensitive, are difficult to distinguish vascular bifurcation point and blood vessel end points simultaneously; 3rd class methods are the bifurcation extracting method based on machine learning algorithm, these class methods mainly utilize machine learning algorithm to build a sorter, train to mark off bifurcation from candidate point to the feature extracted from image (as proper vector histogram, gray-scale value local mean value etc.).These class methods are owing to employing machine learning algorithm, and can take many time in the training stage, extraction efficiency is lower.

Summary of the invention

The object of the embodiment of the present invention is the vascular bifurcation point extracting method providing a kind of lung CT image, is intended to solve existing intersecting blood vessels point extracting method and requires high and to noise-sensitive problem to refinement result.

The embodiment of the present invention is achieved in that a kind of vascular bifurcation point extracting method of lung CT image, said method comprising the steps of:

Bronchus in lung CT image and blood vessel are strengthened, and determines in the pixel after strengthening, belong to the pixel of blood vessel;

Each pixel belonging to blood vessel carries out spheric tensor ballot respectively to other pixel belonging to blood vessel in ballot window;

According to spheric tensor voting results, the spheric tensor poll of every bit accumulation is decomposed, obtains the set of alternative vascular bifurcation point;

Utilize Principal Component Analysis Algorithm to analyze each alternative vascular bifurcation point, to remove the distal point of blood vessel, obtain vascular bifurcation point accurately.

Another object of the embodiment of the present invention is the vascular bifurcation point extraction system providing a kind of lung CT image, and described system comprises:

Blood vessel pixel determination module, for strengthening the bronchus in lung CT image and blood vessel, and determines in the pixel after strengthening, belongs to the pixel of blood vessel;

Spheric tensor vote module, carries out spheric tensor ballot for controlling each pixel belonging to blood vessel respectively to other pixel belonging to blood vessel in ballot window;

Alternative vascular bifurcation point set acquisition module, for according to spheric tensor voting results, decomposes the spheric tensor poll of every bit accumulation, obtains the set of alternative vascular bifurcation point;

Vascular bifurcation point extraction module, for utilizing Principal Component Analysis Algorithm to analyze each alternative vascular bifurcation point, to remove the distal point of blood vessel, obtains vascular bifurcation point accurately.

The vascular bifurcation point extracting method of lung CT image that the present invention proposes and system take full advantage of the additive of the ball-type feature of vascular bifurcation point on geometry and Tensor Voting, by carrying out spheric tensor ballot to the pixel lung CT image after enhancing belonging to blood vessel, solve the local maximum of spheric tensor conspicuousness coefficient to obtain alternative vascular bifurcation point, utilize Principal Component Analysis Algorithm to remove the distal point of blood vessel afterwards, obtain pulmonary vascular bifurcation accurately.The method and system execution efficiency higher, without the need to image preceding processing operations, to noise, there is stronger robustness, can estimate to provide important evidence for the image registration in later stage and pulmonary movements, be conducive to the lung CT image analysis in later stage, adjuvant clinical Clinics and Practices.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the vascular bifurcation point extracting method of the lung CT image that the embodiment of the present invention provides;

Fig. 2 is in the embodiment of the present invention, determines the first detail flowchart of the pixel belonging to blood vessel;

Fig. 3 is in the embodiment of the present invention, determines the second detail flowchart of the pixel belonging to blood vessel;

Fig. 4 is in the embodiment of the present invention, to the detail flowchart that eigenwert and proper vector are reconstructed;

Fig. 5 is in the embodiment of the present invention, determines the detail flowchart of vascular bifurcation point accurately;

Fig. 6 is the structural drawing of the vascular bifurcation point extraction system of the lung CT image that the embodiment of the present invention provides;

Fig. 7 is the first structural drawing of Fig. 6 medium vessels pixel determination module;

Fig. 8 is the second structural drawing of Fig. 6 medium vessels pixel determination module;

Fig. 9 is the structural drawing reconstructing submodule in Fig. 8;

Figure 10 is the structural drawing of Fig. 6 medium vessels bifurcation extraction module.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

The vascular bifurcation point extracting method of lung CT image that the present invention proposes takes full advantage of the additive of the ball-type feature of vascular bifurcation point on geometry and Tensor Voting, by carrying out spheric tensor ballot to the pixel lung CT image after enhancing belonging to blood vessel, solve the local maximum of spheric tensor conspicuousness coefficient to obtain alternative vascular bifurcation point, utilize Principal Component Analysis Algorithm to remove the distal point of blood vessel afterwards, obtain pulmonary vascular bifurcation accurately.

Fig. 1 shows the flow process of the vascular bifurcation point extracting method of the lung CT image that the embodiment of the present invention provides, and comprises the following steps:

S1: the bronchus in lung CT image and blood vessel are strengthened, and determine in the pixel after strengthening, belong to the pixel of blood vessel.

As the first preferred implementation, as shown in Figure 2, step S1 is further comprising the steps:

S11: utilize multiple dimensioned Gaussian function smoothing to lung CT image.

Suppose G (x, y, z; σ) be the three-dimensional Gaussian function that yardstick is σ, then the sharpening result I of lung CT image I (x, y, z) under yardstick σ _σ(x, y, z) is expressed as: wherein, $G(x,y,z;\mathrm{\σ})=\frac{1}{{\left(\sqrt{2\mathrm{\π}}{\mathrm{\σ}}^2\right)}^3}{e}^{-\frac{x^2+y^2+z^2}{2{\mathrm{\σ}}^2}}.$

S12: under each yardstick, calculates the Hessian matrix of every bit in lung CT image according to sharpening result.

Under supposing yardstick σ, the Hessian matrix at lung CT image mid point (x, y, z) place is H _σ(x, y, z), then it is expressed as:

$H_{\mathrm{\σ}}(x,y,z)=\left[\begin{array}{ccc}\frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{{\∂x}^2}& \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{\∂x\∂y}& \frac{\∂I_{\mathrm{\σ}}^2(x,y,z)}{\∂x\∂z}\\ \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{\∂x\∂y}& \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{{\∂y}^2}& \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{\∂y\∂z}\\ \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{\∂x\∂z}& \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{\∂y\∂z}& \frac{{\∂I}_{\mathrm{\σ}}^2(x,y,z)}{{\∂z}^2}\end{array}\right]$

S13: carry out Eigenvalues Decomposition to the Hessian matrix of every bit, obtains three eigenwerts and distinguishes proper vector one to one with three eigenwerts.

In the embodiment of the present invention, for Hessian matrix H _σthree eigenwerts that (x, y, z) obtains after decomposing are designated as λ ₁, λ ₂, λ ₃, and meet | λ ₁|≤| λ ₂|≤| λ ₃|; With three eigenwerts respectively one to one three proper vectors be designated as

The eigenwert of Hessian matrix and proper vector can describe the geometric properties of tubular structure.Specifically, for the point belonged on tubular structure, its proper vector characteristic of correspondence value along blood vessel trend is one less in three eigenwerts; And edge and blood vessel move towards all the other two proper vector Zhang Chengyi planes in vertical tangent plane direction, and all the other two proper vector characteristic of correspondence value sizes are close, and are two larger in three eigenwerts, namely meet | λ ₃| ≈ | λ ₂| " | λ ₁| ≈ 0.

S14: according to eigenwert and the proper vector of every bit, estimates that corresponding point belong to the possibility of tubular structure under each yardstick.

Under supposing yardstick σ, three eigenwerts that image mid point (x, y, z) is corresponding meet | λ ₁|≤| λ ₂|≤| λ ₃|, the possibility that point (x, y, z) belongs to tubular structure is V _s(σ), then meet:

$V_s\left(\mathrm{\σ}\right)=\left\{\begin{array}{cc}0& {\mathrm{\λ}}_2>0\mathrm{or}{\mathrm{\λ}}_3>0\\ (1-e^{-\frac{R_A^2}{2{\mathrm{\α}}^2}})\·e^{-\frac{R_B^2}{2{\mathrm{\β}}^2}}\·(1-e^{-\frac{S^2}{2{\mathrm{\γ}}^2}})\·e^{-\frac{2{\mathrm{Coeff}}^2}{\left|{\mathrm{\λ}}_2\right|{\mathrm{\λ}}_3^2}}& \mathrm{otherwise}\end{array}\right.$

Wherein, $R_A=\frac{\left|{\mathrm{\λ}}_2\right|}{\left|{\mathrm{\λ}}_3\right|},$ $R_B=\frac{\left|{\mathrm{\λ}}_1\right|}{\sqrt{\left|{\mathrm{\λ}}_2{\mathrm{\λ}}_3\right|}},$ $S=\sqrt{{\mathrm{\λ}}_1^2+{\mathrm{\λ}}_2^2+{\mathrm{\λ}}_3^2},$ Coeff is a constant, and α is constant, and generally desirable 0.5, β is constant, and generally desirable 0.5, γ is the constant set.

S15: the maximal value of getting the possibility of every bit under different scale belongs to the end value of the possibility of tubular structure as corresponding point.

Suppose that the end value of this possibility is designated as V, then have: wherein, σ _min, σ _maxsmallest dimension and out to out respectively.

S16: utilize disperse function to upgrade the intensity that each possibility in lung CT image is greater than the point of 0, until update times reaches maximum iteration time, the blood vessel intensity after being enhanced.

In the embodiment of the present invention, utilize VED algorithm to carry out disperse to the possibility of tubular structure in figure, disperse function representation is: wherein, V _tbe the tubular structure intensity after disperse, t is the disperse time, ▽. be divergence operator, D is dispersion tensor, and meets:

${\mathrm{\λ}}_1^{\′}=-+(\mathrm{\ω}-1)V^{\frac{1}{L}}$

${\mathrm{\λ}}_2^{\′}={\mathrm{\λ}}_3^{\′}=1+(\mathrm{\ϵ}-1)V^{\frac{1}{L}}$

Wherein, for Hessian matrix H _σ(x, y, z) decompose after three eigenwerts obtaining, ω is a parameter, in order to show the intensity of anisotropy disperse, desirable ω=5, ε is a parameter, in order to ensure that dispersion tensor D is a positive definite matrix, desirable ε=0.01, L is a parameter, in order to control the susceptibility that disperse function affects blood vessel, and desirable L=2.

S17: judge strengthen after blood vessel intensity reach set threshold value point as belonging to the pixel of blood vessel.

Suppose that setting threshold value is Th, if in the result after strengthening, the blood vessel intensity V (x, y, z) after the enhancing of point (x, y, z) meets V (x, y, z) >=Th, then decision-point (x, y, z) is for belonging to the pixel of blood vessel.Wherein, the mean value of the blood vessel intensity in the lung CT image after the desirable enhancing of threshold value Th after each point enhancing is set.

As the preferred implementation of the second, as shown in Figure 3, different from shown in Fig. 2, now, after the end value being obtained possibility by step S15, before step S16, further comprising the steps of:

S18: be greater than the minimum direction of eigenwert of the point of 0 with possibility for normal direction, the point other possibility in neighborhood being greater than to 0 carries out excellent Tensor Voting, and according to voting results, each possibility is greater than to the eigenwert of the point of 0 and proper vector is reconstructed, move towards direction with what determine that each possibility is greater than the tubular structure of the point of 0.

Further, as shown in Figure 4, step S18 can comprise the following steps again:

S181: with the point that each possibility is greater than 0 be polling place, with the minimum direction of the eigenwert of corresponding polling place for normal direction, the point other possibility in neighborhood being greater than to 0 is that poll acceptance point carries out excellent Tensor Voting.

Suppose that three eigenwerts corresponding to point (x, y, z) that possibility is greater than 0 meet | λ ₁|≤| λ ₂|≤| λ ₃|, proper vector is rod tensor is S, and plate tensor is P, and spheric tensor is B, then the Hessian matrix H of point (x, y, z) under corresponding yardstick can be analyzed to excellent tensor, plate tensor sum spheric tensor sum, namely has: H=(λ ₃-λ ₂) S+ (λ ₂-λ ₁) P+ λ ₁b, wherein, (λ ₃-λ ₂) represent curved-surface display.

In embodiments of the present invention, suppose that possibility is greater than the point (x, y, z) of 0 for excellent Tensor Voting point, with the direction that the eigenwert of point (x, y, z) is minimum for normal direction, the some R other possibilities in neighborhood being greater than to 0 votes, and R is poll acceptance point, then point (x, y, z) is the excellent tensor Stick (l, θ, π) comprising direction and intensity to the poll that a R launches, and meets:

$\mathrm{Stick}(l,\mathrm{\θ},\mathrm{\π})=({\mathrm{\λ}}_3-{\mathrm{\λ}}_2)\mathrm{DF}(s,k,\mathrm{\σ})\left[\begin{array}{c}-\sin \left(2\mathrm{\θ}\right)\\ \cos \left(2\mathrm{\θ}\right)\\ 0\end{array}\right]\left[\begin{array}{ccc}-\sin \left(2\mathrm{\θ}\right)& \cos \left(2\mathrm{\θ}\right)& 0\end{array}\right]$

Wherein, $\mathrm{DF}(s,k,\mathrm{\σ})=e^{-\frac{s^2+c{k}^2}{{\mathrm{\σ}}^2}}$ For conspicuousness attenuation function, and $s=\frac{\mathrm{\θl}}{\sin \mathrm{\θ}},$ $k=\frac{2\sin \mathrm{\θ}}{l},$ θ be point (x, y, z) with the line l of some R with the plane included angle of opening, the normal direction of the plane of opening is s is the arc length of line l, and σ specifies the range scale of ballot, and determine the size of ballot window, c is the function of range scale σ, for restricting the degree of degeneration of curvature, and meets:

In the embodiment of the present invention, by vectorial for normal direction with minimal eigenvalue character pair, add again an excellent tensor conspicuousness (λ ₃-λ ₂) as weight, carry out excellent Tensor Voting.After poll closing, ballots that each possibility in image is greater than capital acquisition surrounding neighbors other points interior of 0 add up.

S182: the eigenwert and the proper vector that according to voting results, each possibility are greater than to the point of 0 are reconstructed, moves towards direction with what determine that each possibility is greater than the tubular structure of the point of 0.

In the embodiment of the present invention, add up to the poll Stick (l, θ, π) that poll acceptance point R place receives, cumulative process comprises the cumulative of tensor size and Orientation, note T _rthe cumulative tensor that ' (x, y, z) receives for acceptance point, carries out feature decomposition to it:

Wherein | λ ' ₃|≤| λ ' ₂|≤| λ ' ₁| be T _rthe eigenwert of ' (x, y, z), for tensor T cumulative after poll closing _rthe proper vector of ' (x, y, z), these three new feature vectors minimum, secondary little, maximum eigenwert of character pair value respectively, the now direction of the proper vector of gained namely be the correction direction to former figure dispersal direction.Now, step S16 is the proper vector according to reconstruct replace original proper vector the blood vessel intensity of disperse function to blood vessel structure is utilized to carry out disperse.

Relative to the first implementation, the second implementation utilizes the tensor direction of reconstruct can reduce the disperse of blood vessel intensity along blood vessel tangent plane preferably, strengthens the disperse along vessel directions simultaneously, reaches restraint speckle, strengthens the effect of blood vessel feature.

S2: each pixel belonging to blood vessel carries out spheric tensor ballot respectively to other pixel belonging to blood vessel in ballot window.

In embodiments of the present invention, the pixel (x, y, z) supposing to belong to blood vessel is spheric tensor polling place, and vote to the some Q in ballot window, Q is poll acceptance point.The smooth sphere of selected tie point (x, y, z) and some Q, note sphere arc is s, then point (x, y, z) is excellent tensor Stick (l to the poll that a Q launches, θ, π) the spheric tensor B that obtains after both direction rotates, and meet:

$B(x,y,z)={\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3{I}_{\mathrm{\σ}}(x,y,z){\∫}_0^{2\mathrm{\π}}{\∫}_0^{2\mathrm{\π}}{S}_{\mathrm{\θ\φ\ψ}}\mathrm{d\φd\ψ}{|}_{\mathrm{\θ}=0}$

Wherein, I _σ(x, y, z) represents the sharpening result of described lung CT image under yardstick σ, S _{θ φ ψ}represent excellent tensor Stick (l, θ, π) result after anglec of rotation φ and angle ψ, angle φ is excellent tensor Stick (l, θ, π) in xy plane, take z-axis as the anglec of rotation of rotation center, angle ψ is excellent tensor Stick (l, θ, π) in xz plane, take y-axis as the anglec of rotation of rotation center, rod tensor Stick (l, θ, π) meets:

$\mathrm{Stick}(l,\mathrm{\θ},\mathrm{\π})=\mathrm{DF}(s,k,\mathrm{\σ})\left[\begin{array}{c}-\sin \left(2\mathrm{\θ}\right)\\ \cos \left(2\mathrm{\θ}\right)\\ 0\end{array}\right]\left[\begin{array}{ccc}-\sin \left(2\mathrm{\θ}\right)& \cos \left(2\mathrm{\θ}\right)& 0\end{array}\right]$

Wherein, $\mathrm{DF}(s,k,\mathrm{\σ})=e^{-\frac{s^2+c{k}^2}{{\mathrm{\σ}}^2}}$ For conspicuousness attenuation function, and $s=\frac{\mathrm{\θl}}{\sin \mathrm{\θ}},$ $k=\frac{2\sin \mathrm{\θ}}{l},$ θ be point (x, y, z) with the line l of some Q with or the plane included angle of opening, l is point (x, y, z) and the line putting Q, s is the arc length of l, and σ is aforesaid yardstick, which specify the range scale of ballot, determines the size of ballot window, c is the function of range scale σ, for restricting the degree of degeneration of curvature, and meets: $c=\frac{-16\log \left(0.1\right)\×(\mathrm{\σ}-1)}{{\mathrm{\π}}^2}.$

In the embodiment of the present invention, the square of ballot window can be a length of side be ws, and length of side ws meets: λ ₁λ ₂λ ₃as the weight of poll, be that the product of three eigenwerts is also larger because three eigenwerts of spherical geometric consequence are all comparatively large.

The spheric tensor poll B that pixel Q obtains _qbe described pixel Q place ballot window in all pixels belonging to blood vessel spheric tensor sum of launching to pixel Q, that is:

S3: according to spheric tensor voting results, decomposes the spheric tensor poll of every bit accumulation, obtains the set of alternative vascular bifurcation point.

In the embodiment of the present invention, step S3 can comprise the following steps again: after spheric tensor poll closing, carries out Eigenvalues Decomposition to the spheric tensor voting results B that each pixel (x, y, z) belonging to blood vessel obtains:

Wherein, λ ₁", λ ₂", λ ₃" be respectively three eigenwerts after Eigenvalues Decomposition, and have | λ ₁"≤| λ ₂" |≤| λ ₃" |, be respectively and described three eigenwerts difference three proper vectors one to one;

If the minimal eigenvalue of point (x, y, z) | λ ₁" | be the maximal value in ballot window, then determine that point (x, y, z) is for alternative vascular bifurcation point.

S4: utilize principal component analysis (PCA) (Principle Component Analysis, PCA) algorithm to analyze each alternative vascular bifurcation point, to remove the distal point of blood vessel, obtain vascular bifurcation point accurately.

Due in the alternative vascular bifurcation point utilizing step S3 to obtain, there is the distal point of blood vessel, utilize PCA algorithm to screen alternative vascular bifurcation point, to obtain vascular bifurcation point more accurately.Now, further, as shown in Figure 5, step S4 can comprise the following steps again:

S41: centered by alternative vascular bifurcation point, gets all coordinates belonging to the pixel of blood vessel in a ballot window, forms a set.

S42: calculate the coordinate average in this set, and covariance matrix.

Assumption set X is X={ (x _i, y _i, z _i) | i=1 ..., N}, wherein, N is the coordinate number belonging to the pixel of blood vessel in ballot window, and coordinate average is (v _x, v _y, v _z), then have if covariance matrix is C, then have:

$C=\frac{1}{N}\left[\begin{array}{ccc}\underset{i}{\mathrm{\Σ}}(x_i-v_x)(x_i-v_x)& \underset{i}{\mathrm{\Σ}}(x_i-v_x)(y_i-v_y)& \underset{i}{\mathrm{\Σ}}(x_i-v_x)(z_i-v_z)\\ \underset{i}{\mathrm{\Σ}}(y_i-v_y)(x_i-v_x)& \underset{i}{\mathrm{\Σ}}(y_i-v_y)(y_i-v_y)& \underset{i}{\mathrm{\Σ}}(y_i-v_y)(z_i-v_z)\\ \underset{i}{\mathrm{\Σ}}(z_i-v_z)(x_i-v_x)& \underset{i}{\mathrm{\Σ}}(z_i-v_z)(y_i-v_y)& \underset{i}{\mathrm{\Σ}}(z_i-v_z)(z_i-v_z)\end{array}\right]$

S43: Eigenvalues Decomposition is carried out to covariance matrix C:

Wherein, λ _{1, C}, λ _{2, C}, λ _{3, C}be respectively three eigenwerts after Eigenvalues Decomposition, and meet λ _{1, C}|≤| λ _{2, C}|≤| λ _{3, C}|, be respectively and three eigenwerts difference, three proper vectors one to one, and get eigenvalue of maximum | λ _{3, C}| characteristic of correspondence vector as principal direction.

S44: with principal direction for normal, excessively alternative vascular bifurcation point make a plane, statistics is in the pixel number belonging to blood vessel of this plane side and is in the pixel number belonging to blood vessel of this plane opposite side, according to the ratio belonging to the pixel number of blood vessel being in these plane both sides, distinguish distal point and the vascular bifurcation point of blood vessel.

In the embodiment of the present invention, suppose with principal direction for normal, excessively alternative vascular bifurcation point (x, y, z) make a plane, the pixel number belonging to blood vessel being in this plane side is N ₁, the pixel number belonging to blood vessel being in this plane opposite side is N ₂, then when or time, the distal point that judgement vascular bifurcation point (x, y, z) is blood vessel instead of vascular bifurcation point.

Fig. 6 shows the structure of the vascular bifurcation point extraction system of the lung CT image that the embodiment of the present invention provides, for convenience of explanation, illustrate only the part relevant to the embodiment of the present invention, this system can be the combination of hardware cell, software unit or the software and hardware unit be built in other all kinds of image transformation system.

Particularly, the vascular bifurcation point extraction system of the lung CT image that the embodiment of the present invention provides comprises: blood vessel pixel determination module 1, for strengthening the bronchus in lung CT image and blood vessel, and determine in the pixel after strengthening, belong to the pixel of blood vessel; Spheric tensor vote module 2, carries out spheric tensor ballot for controlling each pixel belonging to blood vessel respectively to other pixel belonging to blood vessel in ballot window; Alternative vascular bifurcation point set acquisition module 3, for according to spheric tensor voting results, decomposes the spheric tensor poll of every bit accumulation, obtains the set of alternative vascular bifurcation point; Vascular bifurcation point extraction module 4, for utilizing Principal Component Analysis Algorithm to analyze each alternative vascular bifurcation point, to remove the distal point of blood vessel, obtains vascular bifurcation point accurately.Wherein, described in the corresponding step S2 of detailed execution flow process of spheric tensor vote module 2, described in the corresponding step S3 of detailed execution flow process of alternative vascular bifurcation point set acquisition module 3, do not repeat.

Further, as shown in Figure 7, the first structure of blood vessel pixel determination module 1 can comprise: level and smooth submodule 11, for utilizing multiple dimensioned Gaussian function smoothing to lung CT image; First calculating sub module 12, under each yardstick, calculates the Hessian matrix of every bit in lung CT image according to sharpening result; Second calculating sub module 13, for carrying out Eigenvalues Decomposition to the Hessian matrix of every bit, obtaining three eigenwerts and distinguishing proper vector one to one with three eigenwerts; Estimation submodule 14, for according to the eigenwert of every bit and proper vector, estimates that corresponding point belong to the possibility of tubular structure under each yardstick; Value submodule 15, the maximal value for getting the possibility of every bit under different scale to belong to the end value of the possibility of tubular structure as corresponding point; Disperse submodule 16, for utilizing disperse function to upgrade the intensity that each possibility in lung CT image is greater than the point of 0, until update times reaches maximum iteration time, the blood vessel intensity after being enhanced; First decision sub-module 17, for judge strengthen after blood vessel intensity reach set threshold value point as belonging to the pixel of blood vessel.Wherein, described in the corresponding step S11 to step S17 of other detailed execution flow process of level and smooth submodule 11, first calculating sub module 12, second calculating sub module 13, estimation submodule 14, value submodule 15, disperse submodule 16, first decision sub-module 17 points, do not repeat.

The second structure of blood vessel pixel determination module 1 as shown in Figure 8.Synchronous with shown in Fig. 7, the second structure also can comprise: reconstruct submodule 18, the minimum direction of eigenwert that possibility for obtaining with value submodule 15 is greater than the point of 0 is normal direction, the point other possibility in neighborhood being greater than to 0 carries out excellent Tensor Voting, and according to voting results, each possibility is greater than to the eigenwert of the point of 0 and proper vector is reconstructed, move towards direction with what determine that each possibility is greater than the tubular structure of the point of 0.Now, disperse submodule 16 is the proper vectors reconstructed according to reconstruct submodule 18, and utilizing disperse function to be greater than each possibility in lung CT image, the intensity of the point of 0 upgrades.

Further, as shown in Figure 9, reconstruct submodule 18 can comprise: excellent Tensor Voting submodule 181, be polling place for being greater than the point of 0 with each possibility, with the minimum direction of the eigenwert of corresponding polling place for normal direction, the point other possibility in neighborhood being greater than to 0 is that poll acceptance point carries out excellent Tensor Voting; Eigenwert and proper vector reconstruct submodule 182, be reconstructed for the eigenwert of point and proper vector according to voting results, each possibility being greater than to 0, moves towards direction with what determine that each possibility is greater than the tubular structure of the point of 0.Wherein, described in the corresponding step S181 and step S182 of detailed execution flow process of excellent Tensor Voting submodule 181 and eigenwert and proper vector reconstruct submodule 182, do not repeat.

Further, as shown in Figure 10, vascular bifurcation point extraction module 4 can comprise: set obtains submodule 41, for centered by alternative vascular bifurcation point, gets all coordinates belonging to the pixel of blood vessel in a ballot window, forms a set; 3rd calculating sub module 42, for the coordinate average (v in set of computations _x, v _y, v _z), and covariance matrix C; Principal direction obtains submodule 43, for carrying out Eigenvalues Decomposition to covariance matrix C: wherein, λ _{1, C}, λ _{2, C}, λ _{3, C}be respectively three eigenwerts after Eigenvalues Decomposition, and meet | λ _{1, C}|≤| λ _{2, C}|≤| λ _{3, C}|, be respectively and described three eigenwerts difference three proper vectors one to one, and get eigenvalue of maximum | λ _{3, C}| characteristic of correspondence vector as principal direction; Second decision sub-module 44, for principal direction for normal, excessively alternative vascular bifurcation point make a plane, statistics is in the pixel number belonging to blood vessel of this plane side and is in the pixel number belonging to blood vessel of this plane opposite side, according to the ratio belonging to the pixel number of blood vessel being in these plane both sides, distinguish distal point and the vascular bifurcation point of blood vessel.Wherein, described in the corresponding step S42 of detailed execution flow process of the 3rd calculating sub module 42, described in the corresponding step S44 of detailed execution flow process of the second decision sub-module 44, do not repeat.

The vascular bifurcation point extracting method of lung CT image that the embodiment of the present invention proposes and system take full advantage of the additive of the ball-type feature of vascular bifurcation point on geometry and Tensor Voting, by carrying out spheric tensor ballot to the pixel lung CT image after enhancing belonging to blood vessel, solve the local maximum of spheric tensor conspicuousness coefficient to obtain alternative vascular bifurcation point, utilize Principal Component Analysis Algorithm to remove the distal point of blood vessel afterwards, obtain pulmonary vascular bifurcation accurately.The method and system execution efficiency higher, without the need to image preceding processing operations, to noise, there is stronger robustness, can estimate to provide important evidence for the image registration in later stage and pulmonary movements, be conducive to the lung CT image analysis in later stage, adjuvant clinical Clinics and Practices.

One of ordinary skill in the art will appreciate that all or part of step realized in above-described embodiment method is that the hardware that can control to be correlated with by program completes, described program can be stored in a computer read/write memory medium, described storage medium, as ROM/RAM, disk, CD etc.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1. a vascular bifurcation point extracting method for lung CT image, is characterized in that, said method comprising the steps of:

Bronchus in lung CT image and blood vessel are strengthened, and determines in the pixel after strengthening, belong to the pixel of blood vessel;

Each pixel belonging to blood vessel carries out spheric tensor ballot respectively to other pixel belonging to blood vessel in ballot window;

According to spheric tensor voting results, the spheric tensor poll of every bit accumulation is decomposed, obtains the set of alternative vascular bifurcation point;

Utilize Principal Component Analysis Algorithm to analyze each alternative vascular bifurcation point, to remove the distal point of blood vessel, obtain vascular bifurcation point accurately.

2. the vascular bifurcation point extracting method of lung CT image as claimed in claim 1, it is characterized in that, described bronchus in lung CT image and blood vessel to be strengthened, and determine in the pixel after strengthening, the step of the pixel that belongs to blood vessel comprises the following steps again:

Utilize multiple dimensioned Gaussian function smoothing to lung CT image, if the three-dimensional Gaussian function that yardstick is σ is G (x, y, z; σ), described lung CT image is I (x, y, z), and the sharpening result of described lung CT image I (x, y, z) under described yardstick σ is I _σ(x, y, z), then describedly utilize multiple dimensioned Gaussian function to smoothing being expressed as of lung CT image: $I_{\mathrm{\σ}}(x,y,z)=I(x,y,z)\⊗G(x,y,z;\mathrm{\σ}),$ Wherein, $G(x,y,z;\mathrm{\σ})=\frac{1}{{\left(\sqrt{3\mathrm{\π}}\mathrm{\σ}\right)}^3}{e}^{-\frac{x^2+y^2+z^2}{2{\mathrm{\σ}}^2}};$

Under each yardstick, calculate the Hessian matrix of every bit in described lung CT image according to sharpening result;

Eigenvalues Decomposition is carried out to the described Hessian matrix of every bit, obtains three eigenwerts and distinguish proper vector one to one with three eigenwerts;

According to described eigenwert and the proper vector of every bit, estimate that corresponding point belong to the possibility of tubular structure under each yardstick;

The maximal value of getting the possibility of every bit under different scale belongs to the end value of the possibility of tubular structure as corresponding point;

Disperse function is utilized to upgrade the intensity that each possibility in described lung CT image is greater than the point of 0, until update times reaches maximum iteration time, the blood vessel intensity after being enhanced;

Judge point that the blood vessel intensity after described enhancing reaches setting threshold value for described in belong to the pixel of blood vessel.

3. the vascular bifurcation point extracting method of lung CT image as claimed in claim 1, it is characterized in that, described bronchus in lung CT image and blood vessel to be strengthened, and determine in the pixel after strengthening, the step of the pixel that belongs to blood vessel comprises the following steps again:

Utilize multiple dimensioned Gaussian function smoothing to lung CT image, if the three-dimensional Gaussian function that yardstick is σ is G (x, y, z; σ), described lung CT image is I (x, y, z), and the sharpening result of described lung CT image I (x, y, z) under described yardstick σ is I _σ(x, y, z), then describedly utilize multiple dimensioned Gaussian function to smoothing being expressed as of lung CT image: $I_{\mathrm{\σ}}(x,y,z)=I(x,y,z)\⊗G(x,y,z;\mathrm{\σ}),$ Wherein, $G(x,y,z;\mathrm{\σ})=\frac{1}{{\left(\sqrt{3\mathrm{\π}}\mathrm{\σ}\right)}^3}{e}^{-\frac{x^2+y^2+z^2}{2{\mathrm{\σ}}^2}};$

Under each yardstick, calculate the Hessian matrix of every bit in described lung CT image according to sharpening result;

Eigenvalues Decomposition is carried out to the described Hessian matrix of every bit, obtains three eigenwerts and distinguish proper vector one to one with three eigenwerts;

According to described eigenwert and the proper vector of every bit, estimate that corresponding point belong to the possibility of tubular structure under each yardstick;

The maximal value of getting the possibility of every bit under different scale belongs to the end value of the possibility of tubular structure as corresponding point;

The minimum direction of eigenwert of the point of 0 is greater than for normal direction with described possibility, the point other possibility in neighborhood being greater than to 0 carries out excellent Tensor Voting, and according to voting results, each possibility is greater than to the eigenwert of the point of 0 and proper vector is reconstructed, move towards direction with what determine that each possibility described is greater than the tubular structure of the point of 0;

According to the described proper vector of reconstruct, disperse function is utilized to upgrade the intensity that each possibility in described lung CT image is greater than the point of 0, until update times reaches maximum iteration time, the blood vessel intensity after being enhanced;

Judge point that the blood vessel intensity after described enhancing reaches setting threshold value for described in belong to the pixel of blood vessel.

4. the vascular bifurcation point extracting method of lung CT image as claimed in claim 3, it is characterized in that, the described minimum direction of eigenwert being greater than the point of 0 with described possibility is for normal direction, the point other possibility in neighborhood being greater than to 0 carries out excellent Tensor Voting, and according to voting results, each possibility is greater than to the eigenwert of the point of 0 and proper vector is reconstructed, to determine that the step moving towards direction that each possibility described is greater than the tubular structure of the point of 0 comprises the following steps again:

With the point that each possibility is greater than 0 be polling place, with the minimum direction of the eigenwert of corresponding polling place for normal direction, the point other possibility in neighborhood being greater than to 0 is that poll acceptance point carries out excellent Tensor Voting;

The eigenwert and the proper vector that according to voting results, each possibility described are greater than to the point of 0 are reconstructed, and move towards direction with what determine that each possibility described is greater than the tubular structure of the point of 0.

5. the vascular bifurcation point extracting method of lung CT image as claimed in claim 1, is characterized in that, if described in belong to the pixel (x, y, z) of blood vessel corresponding three eigenwerts be λ ₁, λ ₂, λ ₃, and meet | λ ₁|≤| λ ₂|≤| λ ₃|, characteristic of correspondence vector is point (x, y, the z) poll that other pixel Q belonging to blood vessel launches in described ballot window is the spheric tensor B (x, y, z) that excellent tensor Stick (l, θ, π) obtains after both direction rotates, and meets:

$B(x,y,z)={\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3{I}_{\mathrm{\σ}}(x,y,z){\∫}_0^{2\mathrm{\π}}{\∫}_0^{2\mathrm{\π}}{S}_{\mathrm{\θ\φ\ψ}}\mathrm{d\φd\ψ}{|}_{\mathrm{\θ}=0}$

Wherein, I _σ(x, y, z) represents the sharpening result of described lung CT image under yardstick σ, S _{θ φ ψ}represent excellent tensor Stick (l, θ, π) result after anglec of rotation φ and angle ψ, angle φ is excellent tensor Stick (l, θ, π) in xy plane, take z-axis as the anglec of rotation of rotation center, angle ψ is excellent tensor Stick (l, θ, π) in xz plane, take y-axis as the anglec of rotation of rotation center, described excellent tensor Stick (l, θ, π) meets:

$\mathrm{Stick}(l,\mathrm{\θ},\mathrm{\π})=\mathrm{DF}(s,k,\mathrm{\σ})\left[\begin{array}{c}-\sin \left(2\mathrm{\θ}\right)\\ \cos \left(2\mathrm{\θ}\right)\\ 0\end{array}\right]\left[\begin{array}{ccc}-\sin \left(2\mathrm{\θ}\right)& \cos \left(2\mathrm{\θ}\right)& 0\end{array}\right]$

Wherein, $\mathrm{DF}(s,k,\mathrm{\σ})=e^{-\frac{s^2+{\mathrm{ck}}^2}{{\mathrm{\σ}}^2}}$ For conspicuousness attenuation function, and $s=\frac{\mathrm{\θl}}{\sin \mathrm{\θ}},k=\frac{2\sin \mathrm{\θ}}{l},$ θ be point (x, y, z) with the line l of some Q with the plane included angle of opening, l is point (x, y, z) and the line putting Q, and s is the arc length of l, and σ specifies the range scale of ballot, and determine the size of ballot window, c is the function of range scale σ, for restricting the degree of degeneration of curvature, and meets: the square of described ballot window to be a length of side be ws, and length of side ws meets:

The spheric tensor poll B that described pixel Q obtains _qbe described pixel Q place ballot window in all pixels belonging to blood vessel spheric tensor sum of launching to described pixel Q, that is:

6. the vascular bifurcation point extracting method of lung CT image as claimed in claim 1, it is characterized in that, described according to spheric tensor voting results, decompose the spheric tensor poll of every bit accumulation, the step obtaining the set of alternative vascular bifurcation point comprises the following steps again:

Eigenvalues Decomposition is carried out to the spheric tensor voting results B that each pixel (x, y, z) belonging to blood vessel described obtains:

Wherein, λ ₁", λ ₂", λ ₃" be respectively three eigenwerts after Eigenvalues Decomposition, and have | λ ₁" |≤| λ ₂" |≤| λ ₃" |, be respectively and described three eigenwerts difference three proper vectors one to one;

If the minimal eigenvalue of described point (x, y, z) | λ ₁" | be the maximal value in ballot window, then determine that described point (x, y, z) is for described alternative vascular bifurcation point.

7. the vascular bifurcation point extracting method of lung CT image as claimed in claim 1, it is characterized in that, the described Principal Component Analysis Algorithm that utilizes is analyzed each alternative vascular bifurcation point, and to remove the distal point of blood vessel, the step obtaining vascular bifurcation point accurately comprises the following steps again:

Centered by described alternative vascular bifurcation point, get all coordinates belonging to the pixel of blood vessel in a ballot window, form a set;

Calculate the coordinate average (v in described set _x, v _y, v _z), and covariance matrix C;

Eigenvalues Decomposition is carried out to described covariance matrix C:

Wherein, λ _{1, C}, λ _{2, C}, λ _{3, C}be respectively three eigenwerts after Eigenvalues Decomposition, and meet | λ _{1, C}|≤| λ _{2, C}|≤| λ _{3, C}|, be respectively and described three eigenwerts difference three proper vectors one to one, and get eigenvalue of maximum | λ _{3, C}| characteristic of correspondence vector as principal direction;

With described principal direction for normal, excessively described alternative vascular bifurcation point make a plane, statistics is in the pixel number belonging to blood vessel of described plane side and is in the pixel number belonging to blood vessel of described plane opposite side, according to the ratio belonging to the pixel number of blood vessel being in described plane both sides, distinguish distal point and the vascular bifurcation point of blood vessel.

8. a vascular bifurcation point extraction system for lung CT image, it is characterized in that, described system comprises:

Blood vessel pixel determination module, for strengthening the bronchus in lung CT image and blood vessel, and determines in the pixel after strengthening, belongs to the pixel of blood vessel;

Spheric tensor vote module, carries out spheric tensor ballot for controlling each pixel belonging to blood vessel respectively to other pixel belonging to blood vessel in ballot window;

Alternative vascular bifurcation point set acquisition module, for according to spheric tensor voting results, decomposes the spheric tensor poll of every bit accumulation, obtains the set of alternative vascular bifurcation point;

Vascular bifurcation point extraction module, for utilizing Principal Component Analysis Algorithm to analyze each alternative vascular bifurcation point, to remove the distal point of blood vessel, obtains vascular bifurcation point accurately.

9. the vascular bifurcation point extraction system of lung CT image as claimed in claim 8, it is characterized in that, described blood vessel pixel determination module comprises:

Level and smooth submodule, for utilizing multiple dimensioned Gaussian function smoothing to lung CT image, if the three-dimensional Gaussian function that yardstick is σ is G (x, y, z; σ), described lung CT image is I (x, y, z), and the sharpening result of described lung CT image I (x, y, z) under described yardstick σ is I _σ(x, y, z), then describedly utilize multiple dimensioned Gaussian function to smoothing being expressed as of lung CT image: $I_{\mathrm{\σ}}(x,y,z)=I(x,y,z)\⊗G(x,y,z;\mathrm{\σ}),$ Wherein, $G(x,y,z;\mathrm{\σ})=\frac{1}{{\left(\sqrt{3\mathrm{\π}}\mathrm{\σ}\right)}^3}{e}^{-\frac{x^2+y^2+z^2}{2{\mathrm{\σ}}^2}};$

First calculating sub module, under each yardstick, calculates the Hessian matrix of every bit in described lung CT image according to sharpening result;

Second calculating sub module, for carrying out Eigenvalues Decomposition to the described Hessian matrix of every bit, obtaining three eigenwerts and distinguishing proper vector one to one with three eigenwerts;

Estimation submodule, for according to the described eigenwert of every bit and proper vector, estimates that corresponding point belong to the possibility of tubular structure under each yardstick;

Value submodule, the maximal value for getting the possibility of every bit under different scale to belong to the end value of the possibility of tubular structure as corresponding point;

Disperse submodule, for utilizing disperse function to upgrade the intensity that each possibility in described lung CT image is greater than the point of 0, until update times reaches maximum iteration time, the blood vessel intensity after being enhanced;

First decision sub-module, the point reaching setting threshold value for the blood vessel intensity after judging described enhancing for described in belong to the pixel of blood vessel.

10. the vascular bifurcation point extraction system of lung CT image as claimed in claim 8, it is characterized in that, described vascular bifurcation point extraction module comprises:

Set obtains submodule, for centered by described alternative vascular bifurcation point, gets all coordinates belonging to the pixel of blood vessel in a ballot window, forms a set;

3rd calculating sub module, for calculating the coordinate average (v in described set _x, v _y, v _z), and covariance matrix C;

Principal direction obtains submodule, for carrying out Eigenvalues Decomposition to described covariance matrix C:

Wherein, λ _{1, C}, λ _{2, C}, λ _{3, C}be respectively three eigenwerts after Eigenvalues Decomposition, and meet | λ _{1, C}|≤| λ _{2, C}|≤| λ _{3, C}|, be respectively and described three eigenwerts difference three proper vectors one to one, and get eigenvalue of maximum | λ _{3, C}| characteristic of correspondence vector as principal direction;

Second decision sub-module, for described principal direction for normal, excessively described alternative vascular bifurcation point make a plane, statistics is in the pixel number belonging to blood vessel of described plane side and is in the pixel number belonging to blood vessel of described plane opposite side, according to the ratio belonging to the pixel number of blood vessel being in described plane both sides, distinguish distal point and the vascular bifurcation point of blood vessel.